BRIDGING EVIDENCE GAPS IN GERIATRIC CARE: - LARGE LANGUAGE MODELS AND KNOWLEDGE GRAPHS FOR DECISION SUPPORT

Sep 29, 2026

Upload by Aria

Study develops an intelligent decision support system that integrates large language models (LLMs) with knowledge graphs (KGs) to close evidence gaps in geriatric care. It builds a high-quality geriatric symptom knowledge graph from clinical guidelines and longitudinal patient-derived symptom networks, then fine-tunes an LLM with domain knowledge and retrieval-augmented generation to align responses with validated evidence. Results show 87.6% accuracy, higher than KG-free fine-tuned LLM (82.5%) and GPT-4 (78.3%), with F1 85.2% and hallucination rate reduced to 3.4%.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages1

Reading Time3 min

AccessFree to download

System overview

Knowledge graph construction

LLM fine-tuning with RAG

Evaluation results

Innovation in Aging, 2025, Vol. 9, No. S2 657 Finally, an interactive validation mechanism cross-referenced LLM-generated outputs with the KG, reinforcing decision accuracy and minimizing misinformation. Evaluation results demonstrated that the proposed system achieved an accuracy of 87.6%, exceeding that of a domain-specific fine-tuned LLM without KG integration (82.5%) and GPT-4 (78.3%) . In terms of classification performance, the system obtained an F1 score of 85.2%, outperforming the fine-tuned LLM (80.4%) and GPT-4 (74.9%) . Furthermore, the system significantly reduced the hallucination rate to 3.4% and provided rationales with citations in 92.1% of cases, ensuring enhanced transparency and reliability. These findings demonstrated that anchoring AI-generated insights in validated knowledge structures increases trust, optimizes care strategies, and supports healthcare professionals in delivering high-quality, individualized geriatric care. Abstract citation ID: igaf122.2243 BRIDGING EVIDENCE GAPS IN GERIATRIC CARE: LARGE LANGUAGE MODELS AND KNOWLEDGE GRAPHS FOR DECISION SUPPORT Shuqi Yang1, Zheng Zhu1, Shuai Wang2, Jiaqing Wang1, Zongan Huang1, and Tiantian Hu1, 1. Fudan University, Shanghai, Shanghai, China (People’s Republic), 2. Dali University, Dali, Shanghai, China (People’s Republic) The increasing prevalence of complex chronic conditions among older adults has highlighted critical gaps in the application of evidence-based practices in geriatric care. While large language models (LLMs) have demonstrated remarkable potential in medical decision support, they are prone to generating hallucinations. To address these challenges, this study developed an intelligent decision support system that integrated LLMs and knowledge graphs (KGs) to enhance evidence-based practice. The system was constructed through a multi-phase construction process: First, a high-quality geriatric symptom knowledge graph was built using structured evidence sources, including clinical guidelines and symptom networks derived from longitudinal patient data. Next, an LLM was fine-tuned with domain-specific knowledge and optimized using retrieval-augmented generation (RAG) to ensure its responses aligned with validated evidence.

FAQs

What problem does the study target in geriatric care?

It targets gaps in applying evidence-based practices for older adults, where LLMs can produce hallucinations in medical decision support.

How is the knowledge graph built in the proposed system?

A geriatric symptom knowledge graph is built using structured evidence sources, including clinical guidelines and symptom networks derived from longitudinal patient data.

Which evaluation metrics demonstrate the system’s effectiveness?

The system reaches 87.6% accuracy and 85.2% F1, reduces the hallucination rate to 3.4%, and provides citations in 92.1% of cases.

Sep 29, 2026

Upload by Aria